Shreya Shankar

Shreya Shankar

@sh_reya

PhD student in databases focusing on MLOps @Berkeley_EECS @UCBEPIC. Prev. 1st ML engineer @viaduct_ai, research @googlebrain, BS & MS @Stanford CS. She/they.

t.co Joined Dec 2022
107
Threads
225
views
33.9K
Followers
3.3K
Tweets

Threads

Shreya Shankar

@nattyice @drewbanin All of what you say makes a ton of sense. But think the one size fits all DB is an ideal -- in my (limited) experience, the new workload precedes the DB custom...

Shreya Shankar

@vboykis This is super interesting but after thinking on it I'm not sure how much I agree with the compiler and problem audiences. It's like saying spell/grammar check is the first...

Shreya Shankar

I've met a few people who quit ML to go into crypto. But these people were either doing DL research (and hadn't written a paper in a while) or chaining protobufs all day long. http...

Shreya Shankar

@TweetAtAKK I liked this talk (may or may not have crashed it, lol). One of @HamelHusain's great points is the high activation energy to merely get started with TFX, which tends to...

Shreya Shankar

Tired of there being 5+ types of data "shift" that interchange meaning in different ML research papers or Medium posts. E.g. covariate shift, domain shift, concept shift, subpopula...

Shreya Shankar

mlops wrapped 2021 ✨ look how broken it is! https://t.co/GQ3HagEtgB

Shreya Shankar

Meta-thread: a 🧡 on writing technical 🧡s [x] is popular / taking our world by storm. However, [y] is a blocker & hard problem. We came up with [z]. Thread: [only include "pull fi...

Shreya Shankar

IMO there's no substitute MLOps experience for building a pipeline that serves predictions at some endpoint (e.g., REST) and trying to sustain some performance over time. Some poin...

Shreya Shankar

Continuous Integration (CI) & testing for ML pipelines is hard and generally unsolved. I’ve been thinking about this for a while now β€” why it’s important, what it means, why curren...

Shreya Shankar

I've made this mistake many times while monitoring ML outputs. Arbitrary thresholds give arbitrary alerts! Our solution was: find the minimum metric val that yields business retur...

Shreya Shankar

(opinionated) thoughtdump of industry vs academic "mlops" tool-building philosophies (writing this more for myself as a means of clarifying my thoughts, and all of this is merely s...

Shreya Shankar

i guess i'll give my honest review of copilot after using it ~6hr/day for 3 weeks (i switched to vscode ~4 years ago, so no learning curve): 1) my stackoverflow usage is down by ~...